根据粗糙集理论进行BP网络设计研究.pdfVIP

　1999 年 4 月 系统工程理论与实践 第 4 期　 根据粗糙集理论进行BP 网络设计的研究 李永敏　朱善君　陈湘晖　韩曾晋　孙增圻 (清华大学自动化系, 北京 100084) 摘要　提出了一种根据粗糙集理论进行BP 网络设计的方法, 它结合了粗糙集理论的强大的定性分 析能力和BP 网络的准确的逼近能力, 得到一种可理解性好、计算简单、收敛速度快的神经网络模型. 这种神经网络的学习算法的要点是: 应用粗糙集的理论和方法, 从给定学习样本数据中发现一组规 则, 并根据这些规则去建立网络模型中相应的隐层节点; 然后用BP 算法迭代求出网络的参数, 从而完 成网络的设计. 关键词　粗糙集　神经网络　 网络 BP A Rough Set A pp roach to BP N eu ral N etw o rk D esign ing L I Yongm in　ZHU Shan jun　CH EN X ianghu i　HAN Zengjin　SU N Zengqi (D epartm ent of A utom ation , T singhua U niversity, Beijing 100084) 　 Abstract A designing m ethod fo r BP neural netw o rk s based on rough set theo ry is p re sented in th is paper. Rough set theo ry has a pow erful capability fo r qualitative analysis, . w h ile BP neural netw o rk s can app roach mo st p roblem s w ith a m uch satisfying accuracy By com bining tho se advantages of the tw o theo ries, w e can construct a k ind of neural , . netw o rk w ith good understandability simp le computation and exact accuracy T he key idea of the learning algo rithm fo r the p ropo sed neural netw o rk is as fo llow s: find a set of rules from the given training data by using Rough Set theo ry, construct the neurons in the h idden layers acco rding to tho se rules, and then learn the argum ents of the neural